Two- and three-dimensional quantitative image analysis of coronary arteries from hi-resolution histological sections

The study of coronary arteries has evolved from examining gross anatomy and morphology to scrutinizing micro-anatomy and cellular composition. Technological advances such as high-powered digital microscopes and high precision cutting devices have allowed clinicians to examine coronary artery morphology and pathology at micron resolution. Our work explores the composition of normal coronary arteries in order to provide the foundation for further study of remodeled tissue. The first of two coronary arteries was sliced into 442 sections with 4 micron inter-slice spacing. Each slice was stained for elastin and collagen. The second coronary artery was sectioned into 283 slices, also with 4 micron resolution. These slices were stained for cellular nuclei and smooth muscle. High resolution light microscopy was used to image the sections. The data was analyzed for collegen/elastin content and nuclei density, respectively. Processing of this type of data is challenging in the areas of segmentation, visualization and quantification. Segmentation was confounded by variation in image quality as well as complexity of the coronary tissue. These problems were overcome by the development of 'smart' thresholding algorithms for segmentation. In addition, morphology and image statistics were used to further refine the results of the segmentation. Specificity/sensitivity analysis suggests that automatic segmentation can be very effective. 3-D visualization of coronary arteries is challenging due to multiple tissue layers. Methods such as summed voxel projection and maximum intensity projection appear to be effective. Shading methods also provide adequate visualization, however it is important to incorporate combined 2-D and 3-D displays. Surface rendering techniques (e.g. texture mapping) are useful tools for visualizing parametric data. Quantification in three dimensions is simple in practice but appropriate descriptions of these results must be displayed to clinicians in a clear way. Preliminary results are promising, but continued development of algorithms for processing histological data is needed.